“To do so, these products are “trained” on datasets, creating massive issues of bias and copyright infringement, because these generative AIs have to learn to construct whatever they’re generating from somewhere.”
Our reading:
Training generative AI on datasets creates massive issues of bias and copyright infringement.
Assessment: Supported
What happened
Does the evidence support the cause?
Outcome: Supported
The evidence matches what they said.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
Read plainly, the claim is that training generative AI on collected data was causing major bias problems and copyright-infringement disputes. Both were well documented by the time of the post. A Stanford-led study first posted in November 2022 found that ordinary prompts to popular image generators produced racial and gender stereotypes, and that mitigation attempts did not stop this. In early 2023, Getty sued Stability AI in the UK and the US, accusing it of misusing more than 12 million photos for training, and artists filed a class action. A 2025 US ruling found that training Claude on lawfully obtained books was fair use. That counts against a strict legal reading of "infringement," but it does not undo the scale of the disputes he described. "Massive" is subjective.
Caveats
- The result depends on reading "issues of copyright infringement" as serious disputes, not established legal liability.
- The link between the bias and the training data is inferred from the research, not stated as proved in the quoted abstract.
Evidence (4 items)
Supports the claim: arXiv / ACM FAccT 2023 (Bianchi et al.),
We find a broad range of ordinary prompts produce stereotypes, including prompts simply mentioning traits, descriptors, occupations, or objects.
Why it matters: Research first posted in November 2022 found widespread racial and gender stereotyping in popular text-to-image models. That supports the bias half of the claim.
- Where
- Abstract
- Publisher
- arXiv / ACM FAccT 2023 (Bianchi et al.)
- Kind of source
- Peer-reviewed paper
- About events on
- Published
Supports the claim: Reuters (via Insurance Journal),
Getty Images has sued artificial intelligence company Stability AI Inc, accusing it in a lawsuit made public on Monday of misusing more than 12 million Getty photos to train its Stable Diffusion AI image-generation system.
Why it matters: A major copyright lawsuit over training data was filed days before the post. That supports the copyright half of the claim.
- Where
- paragraph 1
- Publisher
- Reuters (via Insurance Journal)
- Kind of source
- News report
- About events on
- Published
Supports the claim: Reuters (via Insurance Journal),
follows a separate Getty case against Stability in the United Kingdom and a related class-action complaint filed by artists in California against Stability and other companies in the fast-growing field of generative AI.
Why it matters: By the time of the post there were several training-data copyright suits in two countries. This shows the disputes were broad, not isolated.
- Where
- paragraph 2
- Publisher
- Reuters (via Insurance Journal)
- Kind of source
- News report
- About events on
- Published
Context only: NBC News,
His decision said Anthropic’s use of the books to train its models, including versions of its flagship AI model Claude, was “exceedingly transformative” enough to fall under fair use.
Why it matters: A 2025 US ruling found that training on lawfully obtained books was fair use, though pirated copies were not. This weakens a reading in which training itself is legally infringement. It falls outside the target period.
- Where
- paragraph 8
- Publisher
- NBC News
- Kind of source
- News report
- About events on
- Published
How we judged it
- Kind of claim
- Causal claim
- Deadline
- None given. We don’t add one.
Where we had to interpret
- Plain reading: training generative AI products (such as ChatGPT and Stable Diffusion) on large collected datasets had, by the time of the post, produced serious, widely recognized problems in two areas: (1) biased or stereotyped outputs, and (2) copyright-infringement disputes over training on other people's work without permission.
- "Issues of copyright infringement" is read as substantial infringement allegations and disputes, such as major lawsuits. It is not read as a final court finding that training is infringement.
- Target period: he describes a current state without naming a period, so the period runs from 2022 (when these products spread widely) to the post date.
Right if: Before the post, credible research documented significant bias in widely used generative models, and major copyright-infringement lawsuits were filed over the use of training data.
Wrong if: No significant documented bias in generative model outputs, and no notable copyright disputes over training data, as of early 2023.
We wrote this test after the outcome was known.
We found 12 claims in this post. See them all.
Full record
Source and dates
Where's Your Ed At, Degenerative AI, paragraph 5.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2023-02-16.
Where it was said (1)
Said once in the posts we read.
: Where's Your Ed At, Degenerative AI
To do so, these products are “trained” on datasets, creating massive issues of bias and copyright infringement, because these generative AIs have to learn to construct whatever they’re generating from somewhere.
- Where
- paragraph 5
- Published
- 2023-02-16
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Supported. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.